Automatic Detection of Toll Tax-exempted Vehicles Using Machine Learning Techniques

Abstract

Roads are essential for connecting cities and easing the movement of people, goods, and services throughout Pakistan. To manage these roads, toll plazas are set up to acquire funds for the construction and maintenance. However, many toll plazas in Pakistan are operated manually. In particular, verification for toll tax-exempted vehicles usually relies on the identification of authorized stickers applied to their windshields. In this research, we first develop a custom dataset of vehicles that contains images of vehicles with stickers and without stickers. In our custom dataset, the vehicles contain two types of stickers; one that makes them exempt from paying the toll tax and other ones, such as stickers of McDonald etc. The dataset is preprocessed to standardized formats, dimensions, and resolution. We perform extensive experiments on the dataset by implementing state-of-the-art machine learning techniques. Experimental results show that YOLOv8 for object detection and ResNet for image classification achieve better results than the existing approaches. YOLOv8 with an accuracy of 91.17% outperformed Faster R-CNN, and RT DETR with 87.37% and 85.00% accuracy, respectively. ResNet with an accuracy of 90.8% outperformed Decision Tree Classifier, and Support Vector Machines with 77.41% and 69.0% accuracy, respectively.